+1, especially the locking evaluation and de-entanglement done by Jim and J-D
as part of 810.
We might be seeing regionserver deadlocks with 0.2.0. See attached partial log,
including stack trace requested from the UI, from a regionserver that reports
to the master but does not handle requests from clients, hanging them. There is
no hint that anything is amiss in the log. All of the IPC handlers are blocked.
Also I'm not sure what to make of the high counts on
CompactSplitThread.compactionQueue :
Thread 9 (regionserver/0:0:0:0:0:0:0:0:60020.compactor):
State: TIMED_WAITING
Blocked count: 217143
Waited count: 393898
Stack:
sun.misc.Unsafe.park(Native Method)
java.util.concurrent.locks.LockSupport.parkNanos (LockSupport.java:198)
java.util.concurrent.locks.AbstractQueuedSynchronizer
$ConditionObject.awaitNanos(AbstractQueuedSynchronizer.java:1963)
java.util.concurrent.LinkedBlockingQueue.poll
(LinkedBlockingQueue.java:395)
org.apache.hadoop.hbase.regionserver.CompactSplitThread.run
(CompactSplitThread.java:76)
but do not think I have enough solid information to file a JIRA and
investigate/fix this yet.
I'm hoping we just won't see this with 0.2.1. :-)
- Andy
On Sun, Aug 31, 2008 at 1:28 AM, stack wrote:
> All known bugs filed against 0.2.1 have been resolved.
> If no objections, I'll cut a release candidate in the
> next day or so. For those interested, here's the list
> of 37 issues fixed since 0.2.0: http://tinyurl.com/6qhehe.
> Thanks.
> St.Ack
>
